Answers in Search of a Question: Proofs of the Tri-

Dimensionality of Space
Craig Callender
Department of Philosophy, University of California, San Diego
La Jolla, CA 92093, U.S.A.
Abstract. From Kants first published work to recent articles in the physics literature,
philosophers and physicists have long sought an answer to the question, why does space
have three dimensions. In this paper, I will flesh out Kants claim with a brief detour
through auss law. I then describe !"chels version of the common argument that
stable orbits are possible only if space is three#dimensional. $fter e%amining ob&ections
by 'ussell and van Fraassen, I develop three original criticisms of my own. (hese
criticisms are relevant to both historical and contemporary proofs of the dimensionality of
space )in particular, a recent one by !urgbacher, F. *+mmer,ahl, C., and -acias.. In
general I argue that modern /proofs0 of the dimensionality of space have gone off track.
ey!or"s1 dimension, space, gravity, Kant, stability, orbits
In Kants first published work, #ho$ghts on the #r$e %stimations of Living &or'es
)2345., he speculated that the dimensionality of space follows from gravitys inverse
square law. (hough he said more about dimensionality, especially in regard to the
phenomenon of handedness, he never developed this line of thought further. !ut the very
idea was revolutionary6 for it is the first time that anyone tackled the question of spatial
dimensionality from a physical perspective, as opposed to a mathematical or conceptual
perspective. Kants con&ecture instigated a long tradition, continuing to this day, of
philosophers and physicists trying to show or e%plain why space is three#dimensional.
(heir efforts e%pand on Kants claim and typically probe areas of physics besides
7ewtonian gravitational theory.
8erhaps the most famous paper in the modern tradition is that by the eminent physicist
9hrenfest. 9hrenfest points out in his brief introduction that the very question /why has
space &ust three dimensions:0 perhaps has /no sense0 and can be /e%posed to &ustified
criticism.0 ;et he says he will not pursue these matters in the paper, and instead leave it
to others to determine /what are the /&ustified0 questions to which our considerations are
fit answers0 )2<23, p. 4==.. (he /answers0 are aspects of physical theory that pick out
three spatial dimensions as singular, such as the oft#heard claim that stable orbits are
possible only in three dimensions. (hough there has been some discussion of the physics
of his /answers,0 there has not been much if any philosophical scrutiny of what the
proper questions are.
In what follows I critically evaluate the argument started by Kant and developed by 8aley
)2>=?., 9hrenfest, and others.
2
$lthough this type of argument has been advanced in
many different areas of physics, the central version threading its way through all of this
history is one started by 8aley. 8aley concludes that there are or must be three spatial
dimensions from an argument including@crucially@the premise that stable orbits are
possible only in three dimensions. I challenge both the truth of this specific premise and
the adequacy of the general argument as an e%planation of dimensionality. (he former
challenge is obviously relevant only to the species of arguments invoking stability and
orbits, whereas the latter is relevant to the whole family of arguments seeking to e%plain
the three#dimensionality of space via some /singular0 feature of physics in three
dimensions. I argue that, viewed as answers to the question that now motivates them,
these answers fail and cannot be saved. 'ather than conclude on a negative point,
however, I would like to suggest that with very different background assumptions )say,
from Kants original perspective or from that of contemporary superstring theory. there
might be something salvageable in these considerations.
1. onte!t
(he idea that there might be more than three spatial dimensions is not very startling to the
contemporary reader. 8rograms in quantum gravity and the popular science literature
abound with speculation that the number of spatial dimensions may be anywhere between
three and twenty#five. Aome have even suggested the number is a fraction or even
comple%, and still others that this number evolves with time.
?
It was not always this way.
(hough higher#dimensional ob&ects were well#known in mathematics@Biophantus and
Ceron of $le%andria discussed ob&ects such as /dynamocubos0, a square multiplied by a
cube@such ob&ects were thought of, at best, as playthings of the imagination. 8appus
)ca. D== $B. counseled against working on such Eimpossible ob&ects, for he didnt think
one should waste time on the impossible. 8appian views dominated thought on
dimensionality and can be found in ancient times through the work of Cermann *ot,e, a
2<
th
century erman philosopher, and even today in van Cleve )2<><.. $gainst this
background it is striking that Kant of all people )given his later views on space and time.
was significantly at odds with the prevailing wisdom of his time. For Kant thought not
only that the inverse square law entailed the three#dimensionality of space, but that od
2
Aee, for instance, !arrow )2<>D., !"chel )2<5D., !urgbacher et al )2<<<., Carnap )2<?4., Caruso and
Favier )2<>3., Cadamard )2<?D., -ariwalla )2<32., -irman )2<>5., 8enney )2<5G., (egmark )2<<3.,
Heyl )2<??., and Hhitrow )2<GG.. For more references and some historical background, see Ianich )2<<?.
and Iammer )2<<D.. 7ot all of these articles use stable planetary orbits in their argument. $mong the
claims on finds in this literature are1 /the hydrogen atom has no bound states for nJD06 /the wave equation
implies that high fidelity communication requires nKD06 /only in nKD do the gravitational, electric and
neutrino fields determine their fields with equal strength06 /atomic spectra imply that nKD06 /neutron
diffraction e%periments imply that nG.0
?
In order1 Leilinger and Avo,il )2<>G., $shmore )2<3D., $rkani#Camed, Cohen, and eorgi )?==2..
?
could have chosen instead an inverse cube law which in turn would have picked out a
four#dimensional space, and so on.
(he twentieth century witnessed remarkable changes in the mathematics, philosophy and
physics of dimensionality. -enger and Mrysohn complete the development of the
topological theory of dimension.
D
Kalu,a and Klein famously use a fifth dimension to
/unify0 gravity and electromagnetism, an attempt motivating contemporary superstring
theories. $nd the philosophers Carnap, 'eichenbach and 8oincarN )who himself made
monumental contributions to the topology of dimensions., among others, advocated
geometric and topological conventionalism and applied it to dimensionality. In none of
these fields was three spatial dimensions metaphysically or conceptually mandated. (he
number three emerged for the conventionalists as simply the most convenient integer for
science to use )where /convenient0 is sometimes read so broadly as to strain the notion..
Once this Econventional choice has been made, 'eichenbach thought, one might still try
to e%plain it with physical arguments of the sort well consider.
E8roofs of the e%istence of three spatial dimensions using stable orbits and the like
persist through these many advances more or less unchanged. If anything, these
advances help this style of argument flourish in the twentieth century and today. Cow
one views these efforts, however, has changed in at least two respects. Buring the
heyday of ordinary language philosophy, the arguments of Kant, 8aley and 9hrenfest are
conceived as showing that the number of spatial dimensions is 'ontingent )see Awinburne
2<5>.. For those interested in conceptual analysis and metaphysical necessity, the idea
that spatial dimensionality varies with physical law is surprising because physical law is
such a weak kind of necessity compared with conceptual or metaphysical necessity.
-eanwhile, for those for whom the number of dimensions is definitely up for grabs, such
as string theorists, the potential surprise would occur if these arguments infused the
number three with any kind of necessity at all.
Indeed, one might ask how is it that the stable orbits tradition has e%isted alongside the
physics of e%tra dimensions, or for that matter, the physics of spacetime. 'egarding the
latter, relativity seems to teach that there is not really space and time, but rather
spacetime. (he question should then be, how many dimensions does spacetime have:
(he question would then be whether the proofs can be e%tended to the relativistic
domain. If so then we have a proof that spacetime is four#dimensional. If relativistic
effects destroy the argument, however, then the most we can conclude is that in the
classical limit spacetime must appear four#dimensional. (he same goes for superstring
theory. 8roofs of the three#dimensionality of space and superstring theory appear to
conflict, one saying there need be three dimensions and the other saying there need be
more. !ut most recent efforts to prove space three#dimensional carefully amend their
claims so as not to rule out superstring theory. (he claim is that physics should appear
three#dimensional at some non#fundamental level, but this leaves open the possibility of
higher dimensions at fundamental levels.
D
For an authoritative discussion of the history of the proof of the invariance of dimension theorem and the
development of the Mrysohn#-enger theories of dimension, see Iohnson )2<3< and 2<>2..
D
". #ant$ %auss an& %ra'ity
It is not hard to fill in the details of what Kant must have had in mind when he said the
inverse square law determined the dimensionality of space to be three. Indeed, it will be
useful for our later discussion to flesh this comment out, noting the connection between
7ewtons inverse square law for gravity, the dimensionality of space, and auss law.
Caving died before auss was born, Kant of course didnt use auss law. 7or did he
need it. -y digression through auss law is motivated by the fact that it &ust falls out
from Kants reasoning and will be important to us later@not by a desire to reconstruct
what Kant actually thought.
auss law is the statement that the total flu% through a closed surface of any shape with
mass )or charge. m inside is equal to the change of net mass )or charge. enclosed
anywhere in that surface, divided by a constant. It is of central importance in many areas
of physics. (o understand the general idea of the law, consider a tube with water flowing
through it. Auppose the water has density P and is traveling with velocity '. Consider an
imaginary surface A cutting the tube and some sub#area of it, Q$. Cow much water will
cross Q$ in time Qt: (he answer is trivial1 the cylinder with si,e vQtQ$ will pass
through Q$, so multiplying this by the density, we have PvQtQ$. If we now divide out
Qt we find the rate of flow, or flu%, of the water1 PvQ$. enerali,ing to surfaces not
normal to the velocity of the water and also summing over all faces Q$ while taking the
limit, we obtain1

S
"A n '
,
where
n '
is the part of the velocity pointing normal to A. If we let ( K P', and A K nA,
where n is normal to the surface, we arrive at the familiar e%pression

S
"A (
for the flu%, where ( and A are vectorial quantities. If we have a faucet with water
flowing out and enclose it within an imaginary surface of arbitrary shape, the amount of
flu% R equals the amount of water crossing this surface. auss law is then the claim that
this flu% is equal to the source times a constant when the surface is closed around the
source. auss law )for water. is intimately bound up with the idea that the amount of
water is conserved, for it ties the amount of water passing through a closed surface to the
changing amount of water inside that surface.
In the case of gravity, there is not some /stuff0 that plays the role of water. He are
instead talking about a vector field and its source is not a faucet but a massive body. (he
7ewtonian gravitational force between two bodies is of course
r (
D
? 2
g
r
m (m

4
where m
2
and m
?
are the masses of the bodies, is 7ewtons constant )with dimensions
of spatial volume, mass and time. and r is the unit distance vector between the two
bodies. He can still define a kind of flu%, though it is no longer a flow in the intuitive
sense. He simply substitute the vector field ( by the gravitational field strength1

S
g "A )
where ) is the gravitational field strength, with )K (
g
Sm. (he gravitational flu% is,
loosely, the amount of gravitational influence a source has. iven one mass a distance r
away from another, ) encodes how much acceleration the force from one body e%erts on
the other. (he flu% R
g
is now the amount of /strength0 passing through the surface. One
then makes a kind of hunch that gravitational flu% will be conserved, which is equivalent
to the idea that 7ewtons law e%presses conservation for the number of field lines
emerging from a mass source. auss law becomes

i
i g
) ( 4
where the sum is over the various mass sources inside the surface and the minus sign
arises from the attractive nature of gravity. (his hunch is spectacularly well confirmed.
(he gravitational version of auss law is crucial to e%plaining a number of observed
effects. (he most familiar are that the force produced by a solid sphere of matter is the
same at the surface of the sphere as it would be if the matter were concentrated at the
source, originally proved by 7ewton. It is also essential in allowing that a spherical shell
of matter produces no gravitational field inside. auss law in both electromagnetism
and gravity is confirmed to an e%tra#ordinary accuracy.
(o see the connections among the force law, auss law and dimensionality, lets look at
an especially simple case. Auppose we have a single massive point source located at p.
9nclose p with two concentric spheres, A
2
and A
?
, centered upon p, as in Figure 2. (he
surface area of such spheres increases as r
?
, of course. Auppose we want flu%
conservation. Cow can we get this if the surface area spreads as r
?
: Aimple1 we demand
that gravitational force is emitted as a spherical e%panding field but that the magnitude of
this field decreases by r
#?
to cancel the increase in area. (hus we need an inverse square
field.
In slightly more detail, the outward flu% from a sphere is


S S
g
gS g"A "A )
)Cere we can use g rather than
n )
, the component of ) normal to the surface A, because
were assuming the field is radial and A is spherical.. 7ow assume that the gravitational
G
acceleration is given by
?
r
g
()

and that the surface area is given by

?
4 r S
. 8lug
in these values into our definition of flu%
() r
r
()
gS g 4 4
?
?

,
_

and we arrive at auss law. (e%tbooks commonly derive the form of the force law from
the assumption of auss law, but as above, it is trivial to derive auss law once one has
the force law. One of course needs various other assumptions6 most notably, that the
force is emitted radially from the source and that space is isotropic. !ut with 9uclidean
space as a kind of background presupposition )which well remove later., isotropy is
implied and radial emission is natural )because of the usual correspondence between
dynamical and spatiotemporal symmetries.. Hith these assumptions in place, one can see
that any field other than an inverse square field would entail that the r
?
s dont cancel and
auss law wouldnt hold. If we had, say, an inverse cube field, then the flu% would
decrease with increasing r )r
#D
r
?
Kr
#2
.. For any field other than an inverse square field the
flu% would depend on the distance of the surface from the source.
Cow do dimensions fit in: In the above reasoning, we implicitly assumed something
about the dimensionality of space, for we held that the force radiates from the source as a
sphere and not a hyper#sphere or a circle, etc. (he source thus radiates into a three#
dimensional space rather than some other space. Iump up one dimension and suppose the
force still emerges from the point radially and into an isotropic space. (he surface area
of the sphere AK4Tr
?

is now replaced with that of the hypersphere A K ?T
?
r
D
. In this case,
however, an inverse square field no longer cancels out the r#dependence and an inverse
cube field would be needed for such a result.

5
Of course, this kind of reasoning must be dramatically complicated when we turn to the
curved geometries and non#trivial topologies allowed by general relativity. 7o longer
can we assume that space is flat and isotropic or that the topology is 9uclidean. $nd no
longer will auss law in general be globally true@a fact related to the lack of global
energy conservation in general relativity. (o see the point quickly, suppose that space
itself is curved as a hyper#sphere. (hen as the force from a mass source spreads, it will
eventually come back on itself1 the lines of force will interfere with one another. In a
variably curved spacetime the lines of force will be enormously complicated. 7o
surprise, then, that auss law is not in general globally true in general relativity.
4
One
can still find constraints on dimensionality in general relativity, but if they are e%act they
will be heavily solution#dependent. 8erhaps one could get something slightly more
general if one imposed restrictions such as that the equations for matter contain only
differentials of second order or less and that the geometry be one with various
symmetries. $lternatively, one might abandon any aspirations for such proofs and
merely view these proofs as relevant in the weak#field appro%imation, where 7ewtonian
theory holds. In this case, one would adopt the same attitude toward general relativity as
modern Eproofs adopt toward superstring theory and the like )mentioned above..
'eturning to Kant and classical physics, we have seen that Kants claim that the
dimensionality of space is related to the inverse square law is, left at that, plainly true.
(hey are intimately connected, once one has the necessary assumptions about forces and
4
For a spherically symmetric distribution of matter, !irkhoff showed that 9insteins equations have a
unique solution, namely, the Achwar,schild solution. $ corollary of this is a version of auss law1 in such
a solution the acceleration of a mass shell in a dust filled universe depends only on the matter inside the
shell.
3
the nature of space in place. Kant himself, in the %stimation, claims a particular direction
of dependency1 that /the three#dimensional character seems to derive from the fact that
substances in the e%isting world act on each other in such a way that the strength of the
action is inversely proportional to the square of the distances.0
G
One might ask why
dimensionality follows from the inverse square law and not vice versa. Hell return to
this question after discussing the modern arguments.
*. The Stable +rbits Ar)ument
Kants suggestion is really &ust that, a suggestion, as opposed to a developed argument for
the three#dimensionality of space. Atarting with 8aley )2>=?., however, others picked up
the idea that a physical argument could be provided for e%plaining the dimensionality.
(hough the physical phenomena vary from argument to argument, by far the most
common is one that begins with our observation of various stable planetary orbits )see,
for instance, !arrow )2<>D., !"chel )2<5D., 9hrenfest )2<23., (angherlini )2<5D.,
(egmark )2<<3., Hhitrow )2<GG... (he idea, in a nutshell, is that the e%istence of stable
planetary orbits e%plains in some sense the three#dimensionality of our world. Hell talk
about what sense of e%planation they are looking for with such proofs in G.?. Cere I &ust
want to present the argument. Its main claim is that stable orbits are possible only )or
nearly only. in three spatial dimensions. 8erfectly circular orbits where the attractive
force is e%actly compensated by the necessary centrifugal force are possible in all spatial
dimensions. !ut circular orbits, the argument goes, are e%tremely unlikely. (he slightest
perturbation destroys the orbit. In our world, we have remarkably stable elliptical orbits.
$ll manner of forces constantly perturb these ob&ects yet still they )for the most part.
continue in elliptical orbits. He can only have these orbits in three dimensions, and this
)somehow. e%plains why there are three dimensions.
In what follows I will focus almost e%clusively on the stable orbits arguments, as I
believe it is representative of the remainder of the arguments and also the most common.
Uirtually every general point I have to make about this argument holds also, so far as I
can see, with the other arguments mentioned in footnote 2 as well. I will follow !"chels
version of the argument because it is a little more sophisticated than some te%tbook
versions of the same claims, but not as cumbersome as 9hrenfests or as quick as
!arrows. *ets now look at !"chels argument in more detail.

He begin with the 8oisson#*a8lace equation for a gravitational potential.
*
+
,
+
,
+
,
+
,
n

+ +

?
?
?
D
?
?
?
?
?
2
?
...
)2.
Cere U is the potential and P is the density of the matter field. 7ote three features of this
equation, all of them important to what follows. First, were assuming it valid in all
spatial dimensions n. Aecond, it is auss law, which we saw above6 that is, this is the
G
(he translation is from an 9nglish translation of the 9stimations forthcoming by 9ric Hatkins. Im
grateful to 8rofessor Hatkins for letting me use an early copy.
>
/differentiable0 form of auss law written for a scalar potential using the identity
, )
. (hird, the electrostatic potential can be described the same way, so &ust as
with Kants argument, what can be said for gravity can be said for electricity )relevant for
the analogous argument using stable atoms..
(he solution of )2. is
?

n
r
C
,
and the force is defined as
2
. ? )

n
r
C n
& .
C is an arbitrary constant and r is the distance to any point in the field. He are thus
assuming that the inverse square force is not true in all spatial dimensions6 rather, in two
dimensions it is inverse r, in three dimensions it is inverse r squared, etc.
For !"chel, a stable orbit is one where r alternates for all time between its perihelion
)minimum. value r
2
and aphelion )ma%imum. value r
?
. Ce considers the two#body
problem, with a body describing a central orbit around another body. *et the mass of the
orbiting body be m. Its angular momentum - is constant and equal to

?
mr )
)?.
where V is the a,imuthal speed.
!"chel then makes the appro%imation that at /e%treme distances from the central body
drSdtK=0 )p. 2??D.. Hhen the velocity vanishes, the kinetic energy becomes
?
?
?mr
)
#
.
Ce then invokes the conservation of energy, (WUKconstant, to write1
?
?
?
?
?
?
2
?
2
?
? ?


n n
r
C
mr
)
r
C
mr
)
.
Msing equation )?., the centripetal force in a circular orbit is
D
?
mr
)
&
'

.
<
He observe that for an eccentric orbit, the attractive force must be less than the
centripetal force at perihelion6 otherwise the planet would crash into the other mass.
Aimilarly, at aphelion, the attractive force must be greater than the centripetal force6
otherwise the planet would escape to infinity. Cence we impose two conditions1 )a. at r
2
,
FXF
c
and )b. at r
?
, FJF
c
. He can then substitute these values into our statement of
conservation, obtaining an inequality because were replacing something small by
something smaller and something large by something larger. -assaging a bit, we obtain1
1
]
1

,
_

<
1
]
1

,
_

2
?
?
?
2
?
2
?
. ? )
?
2
. ? )
?
2
n
mr
)
n
mr
)
.
'emembering that r
?
Jr
2
, this condition rules out any elliptic orbit for nY4.
(he other proofs begin the same way. 9hrenefest begins with the 8oisson equation, finds
the equations of motion, and then looks for tra&ectories for which
r
has real and
alternatively positive and negative values. If this is not the case then the planet must
either crash into the one its orbiting or escape to infinity. *ike !"chel he finds that in
nY4 there can be no stable orbits6 however, unlike !"chel he finds that there can be stable
orbits in nK? but these orbits are not closed and lack other desirable properties. !arrow
)2<>D. instead invokes the following two conditions as necessary without comment1
r
D
F)r. = as r =
r
D
F)r. as r .
(hese can be reproduced from some standard te%tbook calculations.
In sum, assuming that the 8oisson equation is valid in all dimensions and that our world
contains stable orbits, the argument to an inverse square law can be read as an argument
for three spatial dimensions. (angherlini )2<5D. makes the same kind of argument but
with a general relativistic analysis in the Achwar,schild solution )more on which later..
8roponents of the stability argument dont seem terribly worried by n X D. 7egative
dimensions are impossible according to any of the usual ways of understanding either
topological or metrical dimension. Furthermore, there are no orbits in nK2, so the
question is really about e%cluding nK?. (his re&ection of two spatial dimensions is done
in a variety of ways. 9hrenfest and !"chel e%clude it because they want the gravitational
potential to vanish at infinity, but it doesnt in nK?. 9hrenfest also mentions !ertrands
theorem at this point, though he does not spell out its role in the argument. One can
easily see how it could be relevant, however. (his theorem states that bounded
tra&ectories are closed only under a central force when the force law is proportional to the
distance between ob&ects or inversely proportional to its square. $ssuming the orbits
must be closed, this would help us e%clude nK? )though see section G.2.. Hhitrow
)2<GG., Cawking )?==?. and (egmark )2<<3. dismiss nK? because biological organisms
would face all sorts of insurmountable topological problems6 nK? worlds are /too barren
to contain observers0 )(egmark, p. 3=.. (he idea here is an anthropic one that we will
2=
come across again. $nother remark one sees is that there is no gravitational force in
general relativity for nXD, but this assumes were working in a general relativistic setting
rather than the 7ewtonian one in the proof.
,. Pre'ious +b-ections
!efore getting to my own ob&ections, lets look at two others that have been made to the
stability argument.
,.1. .ussells +b-ection
In his dissertation#turned#book, 'ussell holds that there are three a priori principles that
ground geometry. One of them is that there are a finite number of spatial dimensions6 the
e%act number, he thought, is contingent and known a posteriori. In making this claim,
'ussell argues, and van Fraassen )2<>G. echoes )p. 2D5., that there might be a small
inaccuracy in 7ewtons law of gravity that remained undetected, but that a small
inaccuracy in space being three#dimensional wouldnt go undetected.
(he limitation of the dimensions to three isZempirical6 nevertheless, it is not
liable to the inaccuracy and uncertainty which usually belong to empirical
knowledge. For the alternatives which logic leaves to sense are discrete@if the
dimensions are not three, they must be two or four or some other number@so that
small numbers are out of the question. Cence the final certainty of the a%iom of
three dimensions, though in part due to e%perience, is of quite a different order
from that of )say. the law of gravitation. )'ussell 2<>3, p. 25D.
From this argument we are supposed to conclude that Kants argument and those like it
cannot succeed.
(hough I feel I understand what 'ussell might have tried to get at with this ob&ection@I
think he is reaching toward an ob&ection I mount later@as it is stated I cant understand
it. (rue, , 7ewtons constant, might differ slightly from what we think it is, or r
?
might
really be r
?.====2
.
5
(hese variables can take on a value in the range of the real numbers,
whereas the topological or metrical dimensionality can only take on a value in the range
of the natural numbers. !ut if read as the simple claim that were always more certain of
the values of natural#numbered quantities than real#numbered quantities, Im not sure I
see a reason to believe this. I have more reason to think the circumference of a circle
divided by twice the radius is D.242GZ than that the number of species on the earth is G
million. Hhy should the difference between a countable versus an uncountable number
of ways of being wrong show up at the level of epistemic warrant:
5
In 2><4 Call, measuring the power law F r
n
from the e%cess precession of -ercury, found n K
#?.======25. (his would have eliminated the need to account for the motion of -ercury with a new
theory6 however, such a value for n would have been inconsistent with other motions in the solar system.
22
Uan Fraassen interprets 'ussell as making the point that the topological properties like
dimension are more fundamental than the metric properties described by gravitation. Ce
writes
Bimensionality is not a metric but a topological feature of space. Cence, the
features of the physical world pointed out are simply not basic enough to shed
much light on the dimensionality of space. )van Fraassen, 2<>G, p. 2D5.
$gain, at least prima facie, the argument is not compelling. (o be sure, mathematically
the spacetime metric requires a manifold with topological structure6 in a formal sense the
topology is more fundamental than the metric. !ut that does not mean that they are not
on a par physically or metaphysically. (here are many senses of Efundamental, and an
argument is needed to show that were more certain about the more mathematically
fundamental. Coming from van Fraassen interpreting 'ussell, this point is ironic. !oth
philosophers are spacetime relationists, i.e., those wanting to found all spatiotemporal
features of the world on distance relations among events. ;et from this perspective one
usually gets the topology from the distance relations between ob&ects, so the metric is
more basic than the topology.
,." /hy %ra'ity0
Hhy does the gravitational force dictate the dimensionality and not )say. the strong
force: (his is a question $bramenko )2<G>. asks. Indeed, we cannot run the same sort
of argument with the strong force, the force binding the nucleus together that acts on
gluons and quarks. In gravity and electromagnetism, the strength of the field decreases
like r
#?
a distance r away from a charge or mass source. (he strong force, which is
transmitted via eight gluons, does not behave this way. (he force between color charges
)quarks and gluons. does not decrease with distance, nor is color flu% conserved. !ut it is
a fundamental force, too, so why doesnt it have some /say0 in the matter of
dimensionality:
'emember, firstly, that any force law can live in any dimensionality. (here is no
problem with having inverse cube laws operate in a three#dimensional space or inverse#
square laws operate in a thirteen#dimensional space. (he relationship among forces,
strengths and the nature of space &ust isnt as /tidy,0 in the sense that non#radial forces,
non#isotropic space, and so on would be needed. Ao the problem is not that the color
force is picking out one dimensionality and gravity another6 if that were so, we would
have to go with the larger dimensionality so as to /fit0 both forces. (he question is
instead why go with the dimensionality that makes for a tidy package with gravity than
one that makes for a tidy package with the color force:
(o this I can imagine two answers. (he first is that at least in the case of the color force
its relationship with distance is very complicated and its hard to imagine any tidy
relationship with space. (he second is that even if one could devise such a package
9hrenfest and especially Kant might point out a reason to favor the gravity package over
others6 namely, gravity and only gravity is a universal force, whereas the strong
2?
interaction occurs only between colored particles, the electromagnetic interaction occurs
only between electrically charged particles, and the weak interaction occurs only between
quarks, leptons and electroweak gauge bosons. ravity, being the only universal force,
gets to /decide.0
In any case, many similar arguments dont appeal to gravity )cf. fn. 2.. One can imagine
repeating this question about any physical feature picked out as special by one of these
arguments. !ut if the complaint is a general one, namely, that these features are too
/contingent0 to play the role assigned to them, then I think they reduce@like 'ussells@
to an ob&ection well soon meet.
1. +b-ections
1.1 Too Simple
$ ma&or empirical premise of the stable orbits argument is that some actual phenomena
are correctly classified as stable orbits. (he argument has force only if the stability it
refers to is what we actually observe. If the conditions on stability are wildly unrealistic
then the proof may be impeccable, as far as it goes, but be irrelevant to our world. !efore
tackling the argument form itself, I would like to question whether this oft#repeated
claim, namely, that stable orbits are possible only in three dimensions, is even remotely
established.
(here are a number of places where one might worry6 but broadly speaking, the worries
will be that what is meant by Estable orbit is either too strong or too weak. If 8oincarN
recurrence counted as completing an orbit, the criterion would clearly be too weak. !ut
the proof could also be too strong1 for instance, if the proof only worked for strictly
periodic orbits, then the planet -ercurys orbit wouldnt count as a stable orbit, for it
travels a slow precessional ellipse that deviates slightly from elliptical and whose
properties evolve over time. If all the other planets, moons, etc. also had orbits that
evolve like -ercurys over time, but &ust less so, then the argument wouldnt apply to
anything in the actual world. Cence one couldnt appeal to empirically observed stability
in our universe, for none of the paradigms of observed stability would count as stable.
Cow do !"chels proof and the others understand stability: (hey assume stable orbits
are always between r
min
and r
ma%
. Atability amounts to having the properties of
permanence and boundedness1 the orbit is permanently bounded by r
min
and r
ma%
. In the
literature concerned with the stability of the solar system, this conception of stability is
considered absurdly weak, for consistent with this definition of stability an orbit may
e%ecute all manner of weird behavior and still be between r
min
and r
ma%
. For this reason
modern discussion of stability uses conceptions such as *iapunov stability, structural
stability, K$- stability, and so on.
From a modern perspective, one would certainly hope for stronger proofs than !"chels.
For present purposes, however, I am more concerned that even !"chels weak sense of
stability is too strong. Hhy: 'eal orbits arent permanently bounded. In the real world,
2D
presumably everything above the molecular scale that we call an orbit actually had a
beginning. Aince the equations are time reversible invariant, that is equivalent to them
having an end point. Uiewed backwards in time, the actual history is one where each
planet either crashed into another mass or escaped into infinity )or !ig !ang singularity..
!y ruling these out, were ruling out the actual history. One might think we could add
entry or e%it paths onto the orbits between r
min
and r
ma%
, but if the orbits are otherwise
e%actly the same then its clear this violates classical mechanics determinism.
$ll we really have evidence of are orbits of less than 2G billion years, so the assumption
that the orbits are eternally between r
min
and r
ma%
seems too strong. Hhen one considers
dimensions other than three, it is not as if all initial tra&ectories crash into the central
planet or begin their &ourney to infinity right a!ay. For some initial conditions one will
find Eorbits that crash or escape much later. Ao we need some confidence that Emuch
later doesnt equal 2G billion years for some initial conditions. (he proofs dont warrant
this confidence. One can imagine future proofs to the effect that unstable orbits are
generic in the relevant solution space in dimensions apart from three, but the e%isting
proofs are far from establishing this6 and such proofs still would leave open the
possibility that we live in )say. five dimensions but with an atypical initial condition.
*ikewise, one might also worry about any proof invoking !ertrands theorem, as
9hrenfest seems to in ruling out stable orbits in two dimensions. !ertrands theorem is
usually quickly glossed as saying that the only bounded tra&ectories of a central force
problem that are closed are those described by a force that is proportional to the distance
between bodies or a force inversely proportional to it square. (he gloss overstates
matters. !ertrands theorem in fact proves that those two force laws are the only
bounded tra&ectories that close for every initial 'on"ition. !ut there are still an infinite
number of closed tra&ectories with other force laws possible6 its &ust that for these other
force laws the tra&ectories dont stay closed under substitution of arbitrary initial
conditions. $n inverse fifth force with ,ero total energy and non,ero angular momentum
leads to a circle through the area of force, for instance.
3
Aince we certainly dont observe
that the closed orbits in our world are also closed given any initial condition, proofs
crucially invoking !ertrands theorem are again too strong.
Finally, when we move to general relativity and non#9uclidean spacetimes, everything
changes once again@as in the Kantian case. In this case too proofs about stability and
orbits will be solution#dependent. Cere it is important to point out, however, that
(angherlini )2<5D. does indeed e%tend the kind of proof were discussing to a
Achwar,schild solution, the natural solution to use for our Keplerian orbit problem. Of
course, this solution and the corresponding proof crucially rely on the assumptions of a
static and spherically symmetric metric, so the proof is only as good as the solution is an
appro%imation. In fact its a very good appro%imation to certain systems )e.g., -ercury#
Aun system. and a poor one to others )e.g., inside a star, or the global spacetime.. $s
before, if the intention of the proof is to base it on fundamental physics, we would want
to use the metric closest to the actual metric of the actual world. If, by contrast, the
3
(his point is made in a *etter to the 9ditor, by -artin A. (iersten )Ameri'an Jo$rnal of Physi's, 3=, 3,
Iuly ?==?, 554..
24
intention is only to show that in certain regimes the proof obtains, one may be satisfied
with such a proof.
Bespite these misgivings, there are other proofs dealing with other phenomena. $nd we
have not uncovered any in principle reason for thinking future arguments about stable
orbits and dimensionality couldnt be convincing. $nd there really is a difference
between three dimensions and the rest regarding at least the simple stability argument.
Ao let us continue, therefore, under the assumption that physics can indeed show that
stable orbits are possible only in three dimensions. Hhat do we make of the argument in
that case:
1." 2s Tri-Dimensionality 3!plaine&0
*et us begin by immediately distinguishing two questions blurred in most of the spatial
dimensions literature1
)2. Cow many spatial dimensions are there:
and
)?. Hhy are there that many spatial dimensions:
(he first question merely tries to figure out the right number6 the second tries to answer
why its that number. (he second question takes the number of spatial dimensions as a
contingent feature of the universe, one needy of e%planation.
Uirtually every participant in this tradition seeks to answer )?. rather than )2.. (hus
!arrow talks of /e%plaining why the world has three dimensions0 )DD3., Hhitrows very
title asks this question, 8enney hopes to present /a /raison d[tre0 for the observed
dimensionality Z of the physical world0 25=3., and so on. 8robably (angherlini states
the claim clearest among non#philosophers. Ce wants to take the claim E(here shall e%ist
stable orbits, make it a principle of n#dimensional dynamics and gravitation theory, and
then show that /the proposition that space is three#dimensional becomes a theorem, rather
than an a%iom0 )5D<. of such physics. (he idea underlying all of this work is admirable
enough1 take something that seems to be brute or necessary in our theoretical framework
and show instead that it is contingent and its value can be derived from this framework.
(here are many ways of answering question )2.. For instance, one might look to
fundamental physics the way 8oincarN and 'eichenbach do. 8oincarN and 'eichenbach,
being conventionalists, thought that there was no fact of the matter regarding
dimensionality until one had made various definitions and assumptions. One could fit the
phenomena, in principle, in any number of dimensions. (he question is which number is
most convenient for science. !oth answer three, 8oincarN due to interesting group
theoretical considerations and 'eichenbach due to a desire to satisfy the principle of
action by contact, which he felt satisfiable only in three dimensions. )$lbert )2<<5.
recently turns the 'eichenbach argument on its head in the quantum realm.. (oday string
2G
theorists and others are arguing that the most natural dimensionality is 2= or 22, since
these dimensions allow for the e%istence of various symmetries and also allegedly unify
and e%plain the e%istence of the forces. 7ote that one need not be a conventionalist to
make these kinds of arguments. 7aturalness, simplicity, consilience, unification, etc.,
might be marks of truth rather than marks of convenience. 9ven the conventionalists
were not so conventionalist about spatial dimensionality@8oincarN suggests that biology
might Ehard#wire us into believing in three spatial dimensions.
I want to grant that the stability argument and Kants considerations might play some role
in answering question )2.. $fter all, in 9uclidean space at least, we do have a very tidy
set of relationships, having the force e%pand isotropically through the space, flu%
conservation, the orbits being stable under perturbation, etc. Aurely any other force laws,
etc. would greatly complicate our physics. (herefore, this argument may help us
determine which dimensionality is the most natural one given the phenomena and
e%planations of other phenomena.

(he same goes for many of the other arguments.
>
!ut, as 'eichenbach states, /one can proceed, after recogni,ing that the three#
dimensionality is a physical fact, to the question of its e%planation, i.e., one can now
search for a cause of the three#dimensionality of space0 )pp. ?3<#?>=.. He can move
from question )2. to question )?.. !eing no stranger to stability arguments like the one
were considering, 'eichenbach in fact gives it as an e%ample of an answer to )?.1
/Auch a proof might read1 If space has n dimensions, and it is a general law of
nature that the attraction between masses varies inversely with the )n#2.th power
of their distance, then the dimensionality of space must be nKD, since otherwise
the motion of the planets and also the arrangement of the stars would not be
stable0 )'eichenbach, 2<G5, p. ?>=..
(hough unsatisfied with such arguments to date because they were not generally
relativistic he endorses the general line. Ce then comments on this strategy1
/(he three dimensionality would thus be recogni,ed as a logical consequence of
certain fundamental properties of matter, which in turn would have to be accepted
as ultimate facts. $ny other attempt at e%planation would be vain. (he three
dimensionality of space cannot be maintained as an absolute necessity6 it is a
physical fact like any other, and therefore sub&ect to the same kind of e%planation0
)'eichenbach, 2<G5, p. ?>=.
*ets suppose that the arguments succeed in showing that there cant be stable orbits in
nJD6 and furthermore, that the laws weve formulated are the laws for any dimension6 and
furthermore, that stability in this argument is neither too strong nor too weak to count
>
For instance, 9hrenfest pointed out a fascinating connection, namely, that there is a dualism between
rotation and translation in three and only three dimensions1 both are defined via three characteri,ing
numbers. In two dimensions, there is only one type of rotation and two types of translation6 in four
dimensions, there are si% types of rotation and four translations, and so on.. ;et in three dimensions there
are three types of rotation and three types of translation. (his deep geometrical fact has many
repercussions throughout dynamics.
25
known phenomena, like the moons orbit, as stable. Can the stability argument answer
the why question:
(here is a strong feeling@which I think 'ussell, van Fraassen and $bramenko were all
e%pressing@that stability is &ust the wrong kind of feature to use to e%plain why space is
three#dimensional. One wants to parody the argument by appealing to something like the
fact that separating inter#locked three#dimensional iron rings is challenging to human
beings in three dimensions, but it would not be challenging in four dimensions or more1
If space has n dimensions, and it is a general law of nature that closed inter#locked
n#dimensional Erings of matter cannot be separated by creatures such as human
beings, then the dimensionality of space must be nKD, since otherwise separating
inter#locked rings wouldnt be challenging in our world
(he feeling is that stability, like the challenge of separating rings, is simply not a deep
enough feature to e%plain dimensionality6 if anything, these facts are symptoms of the
dimensionality.
*ets look more closely at the argument1
2. Apace has n dimensions
?. It is a general law of nature that the attraction between masses varies inversely
with the )n#2.th power of their distance
D. If proposition ? is the case, then planets are stable only if nKD
4. (he planets are stable
G. (herefore, it isSmust be the case that nKD.
I take it that we want the Emust be reading in G to get an answer to question )?. rather
than )2. at the beginning of this section. (he Eis reading &ust adds stability to the pot of
evidence we have for thinking space is in fact three#dimensional. Atability would give
Cempelian confirmation of the hypothesis that space is three#dimensional, but it wouldnt
e%plain the dimensionality any more than finding a raven that is black e%plains why all
ravens are black. (o get an e%planation we need the Emust reading in G, but for that we
need premise 4 to read that the planets must be stable. In 'eichenbachs terminology, we
need to treat his ultimate facts as absolute necessities6 in (angherlinis terminology, it
must be an a%iom that the planets are stable. (he obvious question is, why treat stability,
which is not a corollary of any fundamental law, as necessary or a%iomatic: Aurely there
didnt have to -e stable orbits.
It is not surprising that at this point one sees appeals to the dubious anthropic argument
made in the literature. Cawking and !arrow, for instance, have provided anthropic
arguments that /e%plain0 the D#dimensionality of space. Cawking, and to some e%tent
(egmark and Hhitrow, have said that life is only compatible with three dimensions, so
the fact that were around now e%plains why there are three dimensions. !ut even if D
spatial dimensions are a necessary condition of life, it doesnt follow that there must be D#
dimensions. It only follows that there are D#dimensions. Mnless one adopts the vaguely
23
theological strong anthropic principle and holds that there must be life, we wont get an
e%planation of why there are three dimensions. Aee Horrall 2<<5 for a critique of
anthropic reasoning in many of its forms.
7ow I suspect that a positivistSempiricist of 'eichenbachs stripe may not be bothered by
this kind of criticism1 he wasnt one of physical necessitys greatest fans and surely didnt
think it necessary for scientific e%planations. In his eyes, such a proof reduces two
phenomena to one phenomenon by linking dimensionality to stable orbits, and that is
presumably all one can hope for. 7or would (angherlini be over#impressed6 for him it is
simply a matter of whether adding Ethere must be stable orbits to the fundamental laws
of physics is worthwhile. Ahould such a claim be a theorem or an a%iom:
(his question is naturally approached via *ewis idea of the !est Aystem. Consider all of
the deductive systems whose theorems are all true. $bout these systems *ewis writes1
Aome are simpler, better systemati,ed than others. Aome are stronger, more
informative than others. (hese virtues compete1 $n uninformative system can be
very simple, an unsystemati,ed compendium of miscellaneous information can be
very informative. (he best system is the one that strikes as good a balance as truth
will allow between simplicity and strength. )*ewis 2<<4, p.43>..
'eichenbachS(angherlinis suggestion, put in this conte%t, is that the proposition that
there e%ist stable orbits is an a%iom of the !est Aystem. (he suggestion is a bit peculiar
since the proposition e%presses a matter of particular fact rather than a generali,ation.
!ut such a novel e%tension is not without precedent.
<

In this case it seems clear, however, that 'eichenbachS(angherlinis suggestion would
not make it into the !est Aystem as an a%iom. Compared to 9insteins equations, Biracs
equation, etc., the claim that there e%ist stable orbits seems very weak. (he proposition
that there are stable orbits is not very strong. $rguably we get the three dimensionality of
space, but that is about it. Furthermore, to actually infer truths about the world from such
a generali,ation it will need to be vastly more complicated than its present form. $fter
all, not &ust anything can get into a stable orbit with anything else. Ao if the
generali,ation is to be true, one will have to restrict its scope quite severely. (his move
will have costs, both in terms of the simplicity of the generali,ation and in terms of its
strength. For these kinds of reason I dont believe stability is powerful or simple enough
a fact to merit inclusion in the !est Aystem. 'ead this way, the stable orbits e%planation
of dimensionality isnt guilty of a fallacy or a serious methodological sin6 its Ewrong
only because its an unfortunate way of organi,ing ones knowledge.
1.* 4o&al 4ayhem: /hat is Physics of 56*0
*et me now turn to what I think is the most fundamental worry about proofs like the one
weve e%amined. In the brief introduction to his 2<23 paper, 9hrenfest asked /Hhat is
<
Following !olt,mann and others I defend such a move for the cosmological low entropy initial condition
required by thermodynamics6 see Callender ?==4.
2>
meant by Ephysics of '
4
or '
3
:0 Ce clearly recogni,ed the importance of the question,
but he left it to others to consider. It is obvious why this question is important. He only
get the result that three dimensions are special if we hold some aspects of our laws of
physics fast and vary the dimensionality. (o say that stable orbits are possible only in
nX4 we need a theory of gravity, etc., in 4, G, 5Z If we allow every physical feature to
vary with the dimensionality@as 8oincarN 2<2D allows@it is hard to see how anything
could turn out physically special. -ost advocates of this type of argument e%plicitly
endorse assumptions about what laws operate in higher dimensions. In fact, !arrow and
(ipler 2<>5 espouse a /principle of similarity01 the assumption that the laws of physics in
worlds of other dimensionality are as similar to ours as possible. (his kind of assumption
finds its way into !"chels proof at the first step1 if we didnt assume the *aplace#
8oisson equation is valid for all n in the proof, it would immediately collapse. (he other
worlds are similar to ours in holding *a8lace#8oisson steady while letting the force law
vary. If instead we assumed that anything goes in each dimension, then we could simply
assume there are stable orbits in any n and then solve for the remaining n#dimensional
laws that allow stable orbits.
'ecent work in physics has more or less proceeded in this fashion without appreciating
the general point. !urgbacher *+mmer,ahl, and -acias 2<<<@henceforth !*-@for
instance, show that there are stable hydrogen atoms in higher dimensions, contrary to the
9hrenfest result against stable hydrogen atoms )see also Caruso and Favier 2<>3..
!*-s argument applies to stable planetary orbits too, an application that they e%plicitly
endorse. *ets review their reasoning, for I think seeing it will make the general lesson
emerge with ease.
Horking with electrodynamics and stable atoms, !*- claim that in all dimensions
stability requires that the electrostatic potential be proportional to r
#2
. (hey -egin with the
assumption that there is stability )described by a potential proportional to r
#2
. in all
dimensions. (hen they work backward, as it were, to discover what laws would be
needed to give r
#2
potentials in all dimensions. For nJD this leads to a modification of
-a%wells equations, yet the solutions of these new equations have the same structure as
for nKD and the force between charges is the same as in nKD. (he main differences,
however, are that the modified -a%well equations do not lead to a aussian law for
charges and in even higher dimensions lead to nonlocal equations.
$s they note, the same reasoning can be used for stable orbits in the case of gravity, and
hence, our case. $gain, they hold fast stable orbits and an r
#2
potential@this time the
gravitational potential. (hey then work backwards and find that they are forced to
modify the 8oisson#*a8lace equation for the 7ewtonian potential in the same way as
they had to modify that equation for the electrostatic potential. $s the 8oisson#*a8lace
equation is simply auss law, auss law is broken in this case too. eneral
relativistically, their changes would entail slightly modifying 9insteins equations in
higher dimensions.
!*- do not deny the validity of the 9hrenfest#style argument. 'ather, they are
essentially making the classic move of turning 9hrenfests mo"$s ponens into a mo"$s
2<
tollens argument. He have an inconsistent set1 \)a. auss lawS8oisson equation, )b.
stability )r
#2
potential., )c. stability in nJD dimensions]. 9hrenfest holds the first two
members fast and denies that there could be stability in nJD dimensions. !*- holds the
last two fast and denies auss law and related higher#level equations. (hey write, /we
think that the specific e%pression for the force between charged particles and the stability
of atoms are of more basic physical importance than the validity of auss law0 )5DD..
(hey do show that auss law is valid for a quantity related to the field strength, but it is
not valid for the field strength itself.
!*- 9hrenfest
$ssume (here is stable motion in nJD auss lawS
described by 2Sr potential 8oisson equation
(hen -odify aussS8oisson equation (here is no stable
motion in nJD
He have a kind of ^uine#Buhem problem in action. Cow are we to arrange these
premises: If we view auss law as basic and important, then we will not be willing to
modify -a%wellsS8oissonsS9insteins equations so as to obtain stable motion. If we are
willing to sacrifice auss law as !*- are, then we can get stable motion. Cow are we
to ad&udicate this debate:
One might think that if the battle were between a fundamental law and a non#fundamental
one, then an argument could be made that the fundamental law should win, that
fundamental laws should not be varied with dimensionality. If one anointed auss law
fundamental, then one would go with 9hrenfest6 if one anointed the form of the potential
fundamental, then one would go with !*-. Hhat is nice about this case is that neither I
nor people Ive talked to about it have any idea which way to go. Clearly !arrow and
(iplers principle of similarity doesnt help here. In the classical case we saw how one
could get the form of the force law )and hence the form of the potential too. from
assuming auss law6 alternatively, we saw how one could get auss law from the form
of the force law )and hence potential.. (he assumptions to get from one to the other are
equally mild, so one lacks strong intuitions about which way to go.
(he problem, I submit, is simply that there is no -a'*gro$n" theoreti'al frame!or* that
can help us answer this question. !oth 9hrenfest and !*- are e%tending some nKD
physical laws to higher dimensions and not e%tending some others. $bsent a developed
physical theory that takes dimensionality as contingent and offers principled physical
constraints on what can happen in different dimensions, there seem to be no standards for
knowing which laws hold in what dimension. Hhich is more fundamental, a r
#2
potential
in higher dimensions )and thus stable orbits there. or auss law in higher dimensions
)and thus no stable orbits there.: (here is no scientific theory of this, and only vague
intuitions fill the vacuum.
?=
Its useful to compare and contrast this case with the following e%ample of an
Ee%planation of a fundamental constant. (here is a school of quantum cosmology that
offer e%planations of why the speed of light is what it is, why _, the critical density of the
universe, is what it is, why neutrino masses are what they are, and so on )see, e.g.,
Uilenkin )2<<G... Cow does one e%plain why ' is D==,===,=== mSs: In short, one posits
an ensemble of worlds, puts a probability metric over all these worlds and then derives
that most of these worlds@or most of the ones with humans in them## have light
traveling close to the speed it travels here. Hildly speculative: ;es. Acratching an itch
to e%plain that would be better off unscratched: ;es. !ut despite these problems, this
e%planation of ' is better than present e%planations of tri#dimensionality in at least one
respect1 at least one is e%plicitly positing a theory in which what we formerly took to be a
fundamental constant is allowed to vary with world and there are clear constraints on this.
!y contrast, 9hrenfest, !"chel, !arrow and (ipler, !*-, and so on, are implicitly
offering theories of higher#dimensional worlds wherein dimensionality is not a constant
but a variable. Mnlike Uilenkin, however, they are not dressing up this theory with
details6 nor do they even acknowledge that this is what they are up to )apart from citing
vague principles like the /principle of similarity0 above.. (o e%plain brute facts in one
theory, one needs another theory. Hithout a developed theory, all we have are our
intuitions and various vague recipes for determining closeness to worlds.
2=

!*- and Caruso and Favier dont learn the lesson of their success. For after critici,ing
the 9hrenfest#style argument, !*- then claim to /have proven by a spectroscopic
e%periment that our space is three#dimensional0 )5D2. and Caruso and Favier claim that
nuclear diffraction e%periments imply that n G. (he idea is that the atomic spectra and
diffraction e%periments, not stability, have implications for the dimensionality of space.
(hey /fi%0 9hrenfests methodology by focusing solely on laws that they believe do not
show /singular aspects0 concerning nKD. (hat is, they look for laws that they believe are
indifferent to the dimensionality of space, e.g., the laws of classical thermodynamics.
(his way, as they see it, they do not build in the answer to their question. (hough
ingenious, the argument falls to the same ob&ection used above against 9hrenfest. (heir
argument is an improvement over 9hrenfests because they seek to use laws and
relationships that do not vary with dimension. !ut still, to get their conclusion, some
physics is of course used, some assumptions are made, and these assumptions may not be
legitimate in higher dimensions, e.g., !*-s assumption that in every dimension the
lowest series contains only transitions with lKkK= or Caruso and Faviers assumption that
classical thermodynamics is valid in all n. In the absence of a successful theory that takes
dimensionality to be variable, there is no way to know if these assumptions are
warranted.
$re there theories that might e%plain the number of dimensions: On the hori,on there
may be. $rkani#Camed, Cohen and eorgi, for instance, propose a dynamical theory of
2=
'ealists about a similarity metric among metaphysically possible but physically impossible worlds may
insist that there is still a fact of the matter about which world is closest to this one. In *ewis system there
presumably would be a fact of the matter at dispute between 9hrenfest and !*-. !ut notice that that
metric is one weighting laws differently than matters of fact, so unless we could determine which is the
fundamental law we will likely be epistemically closed to the solution even if there is a fact of the matter at
issue.
?2
dimension creation that would, I suppose, take dimensionality as something to be
appropriately e%plained via the dynamics of physical processes. !ut such theories are far
too speculative for us to have much faith in today.
7. onclusion: /hat Questions are 89ustifie&:0
9hrenfest wrote that he didnt know if the question, /why does space have three
dimensions:0 is a /&ustified0 question. In the preceding sections I have argued that at
least according to the dominant contemporary way of understanding this question
)question )?. of G.?., it is not a good one to ask. (he problem has been that these proofs
have substituted intuitions about higher#dimensional physics where one needs a theory.
Bespite this problem, some of the /answers0 provided by 9hrenfest and others are quite
interesting. I submit that their interest lies primarily in their relevance to answering
question )2. of G.?, the question of determining what in fact the number of spacetime
dimensions is. 7oticing connections between various disparate phenomena is the
lifeblood of science. (hese connections help us determine the overall best
systemati,ation of the world, part of which is a commitment to a particular
dimensionality.
*et me close by noting how our understanding of the question, why does space have three
dimensions:, depends crucially on what background assumptions are in play. Cere are
two e%amples.
First, return to the instigator of this whole pro&ect, Kant. Kant wrote, /substances in the
e%isting worldZhave essential forces of such a kind that they propagate their effects in
union with each other according to the inverse#square relations of the distances, Z`anda
that the whole to which this gives rise has, by virtue of this law, the property of being
three#dimensional0. Hhy did he think that the laws and forces determine the
dimensionality and not vice versa: (he answer is that early Kant was a kind of spatial
relationist. Ce thought that substances and their forces were primary and that spatial
properties were derivative upon the relationships among the substances and their forces.
From this perspective, Kants gravitational argument and stable orbits arguments are
natural ones to make. $fter all, relationists owe us a story about how the various
spatiotemporal properties@e.g., orientability, dimensionality, etc@are grounded in the
relationist base )whatever it is.. Atepping back from the details of Kants argument, one
can think of an argument like his as an attempt to get a spatiotemporal feature from a
dynamical one, as a way of getting a topological feature of spacetime from dynamical
features of matter. Mnderstood as the question, Ewhy does space have three dimensions,
given relationist basis !:, the question is certainly a good one and Kants answer is not
an unreasonable one.
Aecond, fast#forward roughly DG= years and consider the closely related question1 why
does spacetime seem to have four dimensions: In the mouth of a superstring theorist, or
any other heir of Kalu,a#Klein theory, the question assumes that spacetime is
fundamentally more than four dimensions. Hith these theories as background, the
amended question is again certainly a good one to ask. 'ecall that the original five#
??
dimensional Kalu,a 2<?2 unification of gravity and electromagnetism did not have
anything like what we would now call a /compactification0 mechanism@a process that
in a certain limit appro%imates a four#dimensional spacetime. Kalu,as original model
had little to say to our question. Only later with Klein and then 9instein and !ergmann
did one get the idea of /rolled up0 compactified dimensions )for the original papers, see
$ppelquist 2<>3.. (hese schemes, and others involving ingenious new ways of /hiding0
e%tra dimensions, continue in contemporary physics. (hey are the answers to the new
question. 9hrenfest#like demonstrations of the /singular0 nature of four dimensions are
only relevant in the sense that they motivate providing such a mechanism in the first
place. Auppose it is true that one can only have stable orbits in four#dimensional
spacetime. (hen as we move from below the 8lanck scale to the classical scale, the
stringy analog of matter had better behave more and more classically and its ten#
dimensional space had better look more and more like three dimensions to such matter if
it is to account for stable orbits and the like.
Ac;nowle&)ements. For useful comments on the paper I owe many thanks to 8aul
Churchland, Ionathan Cohen, Ioan -untean, and 9ric Hatkins, two anonymous referees,
and audiences at the Mniversity of Florida and 'utgers Mniversity. For the 9nglish
translation of Kant, Im grateful to 9ric Hatkins, and for Figure 2, Im grateful to !en
-agoos.
.eferences
$bramenko, !. )2<G>.. On the Bimensionality and Continuity of 8hysical Apace
and (ime. #he .ritish Jo$rnal for the Philosophy of S'ien'e, D4, ><#2=<.
$lbert, B. )2<<5.. 9lementary ^uantum -echanics. In Cushing, Fine and oldstein
)eds.. .ohmian )e'hani's/ An Appraisal. Bordrecht1 Kluwer, ?33#?>4.
$ppelquist, (., Chodos, $., and 8..O. Freund. )2<>3.. )o"ern al$0a1lein #heories
)Frontiers in 8hysics, vol. 5G., -enlo 8ark, C$1 $ddison#Hesley.
$rkani#Camed, 7., Cohen, $. and C. eorgi )?==2.. )Be.constructing Bimensions
Physi'al 2evie! Letters >5, 43G3#4352.
$shmore, I.F. )2<3D.. On 'enormali,ation and Comple% Apace#time Bimensions.
Comm$ni'ations in )athemati'al Physi's ?<1233#2>3.
!arrow, B. )2<>D.. Bimensionality. Philosophi'al #ransa'tions of the 2oyal So'iety.
*ondon. Aeries $, D2=, DD3.
!arrow, B. and (ipler, F. )2<>5.. #he Anthropi' Cosmologi'al Prin'iple. O%ford1
Clarendon 8ress.
!"chel, H. )2<5<.. Hhy is Apace (hree#Bimensional: (rans. Ira -. Freeman. Ameri'an
Jo$rnal of Physi's. D3, 2???#2??4.
!urgbacher, F. *+mmer,ahl, C., and -acias, $. )2<<<., Is (here a Atable Cydrogen
$tom in Cigher Bimensions:, Jo$rnal of )athemati'al Physi's, 4=, ?, 5?G#5D4.
Callender, C. )?==4.. /-easures, 9%planation and (ime1 Can He 9%plain the !oundary
Conditions of the Mniverse:0 .ritish Jo$rnal for the Philosophy of S'ien'e, GG,
2<G#?23.
?D
Carnap, '. )2<?4.. Breidimensionalit+t des 'aumes und Kausalit+t. Annalen "er
Philosophie $n" philosophis'hen riti* 4, 2=G#2D=.
Caruso, F. and Favier, '.-. )2<>3.. On the 8hysical 8roblem of Apatial Bimensions1 $n
$lternative 8rocedure to Atability $rguments &$n"amenta S'ientiae >, 3D#<2.
9hrenfest, 8. )2<23. In Hhat Hay Boes It !ecome -anifest in the Fundamental *aws
of 8hysics that Apace Cas (hree Bimensions: Pro'ee"ings of the Amster"am
A'a"emy ?=, ?==#?=<.
iddings and (homas )?==?.. Cigh 9nergy Colliders as !lack Cole Factories1 (he 9nd
of Ahort Bistance 8hysics. Physi'al 2evie! B5G, G5=2=ff.
Cadamard, I. )2<?D.. Le't$res on Ca$'hy3s pro-lem in linear partial "ifferential
e4$ations. 7ew Caven1 ;ale Mniversity 8ress.
Cawking, A. )?==2.. #he Universe in a 5$tshell. 7;1 !antam !ooks.
Iammer, -. )2<<D.. Con'epts of Spa'e/ #he 6istory of #heories of Spa'e in Physi's, D
rd

9dition. 7;1 Bover.
Iohnson, B. -. )2<3<.. (he problem of the invariance of dimension in the growth of
modern topology. I, Ar'hive for 6istory of %+a't S'ien'es "< )?., <3#2>>.
Iohnson, B. -. )2<>2.. (he problem of the invariance of dimension in the growth of
modern topology. II, Ar'hive for 6istory of %+a't S'ien'es "1 )?#D., >G#?53.
Ianich, 8. )2<<?.. %$'li"3s 6eritage/ 7s Spa'e #hree1Dimensional8 Bordrecht1 Kluwer.
Kant, I. )2345.. (e"an*en von "er !ahren S'h9t0$ng "er le-en"igen r9fte $n"
.e$rteil$ng "er .e!eise "erer si'h 6err von Lei-ni0 $n" an"ere )e'hani*er in
"ieser Streitsa'he -e"ient ha-en, ne-st einigen vorhergehen"en .etra'ht$ngen
!el'he "ie raft "er :rper ;-erha$pt -etreffen. ^uotes are from an 9nglish
translation, forthcoming, by 9ric Hatkins.
Kalu,a, (. )2<?2.. Lum Mnit+tsproblem der 8hysik, Sit0.-er. Pre$<. A*a". =iss. .erl.
1,, <55#<3?.
*ewis, B. )2<<4.. Cumean supervenience debugged. )in" 2=D, 43D#4<=.
-irman, '. )2<>5.. ^uantum -echanics Betermines the Bimension of Apace. Annals of
the 5e! >or* A'a"emy of S'ien'es 4>=, 5=2#5=D.
-ariwalla, K. C. )2<32.. Bimensionality of Apace#time. Jo$rnal of )athemati'al
Physi's 2?, <5#<<.
8aley, H. )2>=?.. 5at$ral #heology. 8hiladelphia1 8rinted for I. -organ by C. -a%well.
8enney, '. )2<5G.. On the Bimensionality of the 'eal Horld. Jo$rnal of )athemati'al
Physi's 5, 25=3#2522.
8oincarN, C. )2><>.. On the Foundations of eometry. #he )onist IF, 2, 2#4D.
8oincarN, C. )2<2D.. Dernieres Pensees. 8aris1 Flammarion.
'eichenbach, C. )2<G5.. #he Philosophy of Spa'e an" #ime )9nglish translation.. 7;1
Bover.
'ussell, !. )2><3.. An %ssay on the &o$n"ations of (eometry, Cambridge1 Cambridge
Mniversity 8ress.
Awinburne 'ichard . )2<5>.. Spa'e an" #ime. *ondon1 -acmillan.
(angherlini, F. '. )2<5D.. Achwar,schild Field in n Bimensions and the Bimensionality
of Apace 8roblem. 5$ovo Cimento ?3, 5D5#5G2.
(egmark, -. )2<<3.. On the Bimensionality of Apacetime. Classi'al an" ?$ant$m
(ravity 1,$ *5<b*3G.
Uan Cleve, I. )2<><.. 'ight, *eft, and the Fourth Bimension. in #he Philosophy of
?4
2ight an" Left$ van Cleve, I. and Frederick, '. )eds.. Bordercht1 Kluwer.
van Fraassen, !as C. )2<>G.. An 7ntro"$'tion to the Philosophy of #ime an" Spa'e, ?
nd
ed., Columbia Mniversity 8ress, 2<>G.
Uilenkin, $. )2<<G.. 8redictions from quantum cosmology. Physi'al 2evie! Letters 341
>45b4<.
Heyl, C. )2<??.. Spa'e, time, matter. 7;1 Bover.
Hhitrow, . I. )2<GG.. Hhy 8hysical Apace Cas (hree Bimensions. #he .ritish Jo$rnal
for the Philosophy of S'ien'e 5, 2D#D2.
Horrall, I. )2<<5.. Is the Idea of Acientific 9%planation Mnduly $nthropocentric: (he
*essons of the $nthropic 8rinciple. Dis'$ssion Paper Series, *A9 Centre for the
8hilosophy of 7atural and Aocial Aciences ?GS<5.
Leilinger, $. and Avo,il, K. )2<>G.. -easuring the Bimension of Apace#time. Physi'al
2evie! Letters G4, ?GGD#?GGG.
?G